Re: [PATCH v2] mm/contig_alloc: fix alloc_contig_range when __GFP_COMP and order < MAX_ORDER

[Jinjiang Tu <tujinjiang@huawei.com>]

在 2025/5/28 10:25, Zi Yan 写道:
> On 27 May 2025, at 22:19, Andrew Morton wrote:
>
>> On Mon, 12 May 2025 09:13:31 +0800 Jinjiang Tu <tujinjiang@huawei.com> wrote:
>>
>>>>> The change below makes the code cleaner. Acked-by: Zi Yan
>>>>> <ziy@nvidia.com>
>>>> Jinjiang, can you integrate that into your patch and resend?
>>> Sorry for late reply. I will do it today.
>>>
>> As far as I know, this is still pending.  The patch does address a
>> WARN() and it's cc:stable, so it's rather important.
>>
>> I'll move the v2 patch back into mm-new and shall keep it there to
>> remind me to bug people about it ;)
>>
>>
>> David (and everyone!), please don't be afraid to grab someone else's
>> patch, apply changes and send it in.  Stuff happens, and doing this is well
>> within accepted processes.
> Hi Andrew,
>
> Based on my understanding, this patch is fixing an out-of-tree driver
> issue, so Jinjiang is responsible to resolve David’s concerns.
>
> Hi Jinjiang,
>
> Let me know if I miss anything. And if you are not planning to keep
> working on it, please let Andrew know so he can drop it.

Yes, this issue doesn't exist for any in-tree users. I found v2 patch is buggy
and reply in the v1 patch (https://lore.kernel.org/all/80039e40-a518-a85b-b955-96fb048a2dd0@huawei.com/). The
fixed patch is little complicated.

Since there is no in-tree user currently, it is enough if we simply fail if the range order < MAX_ORDER with __GFP_COMP.

If it is valuable to handle this situation, in case in-tree users in the future need to allocate folio with order < MAX_ORDER,
I would like to keep working on it.

> Thanks.
>
> Best Regards,
> Yan, Zi